← back to catalog · registered 2026-11-11 21:03

leonice2/Qwen3.8-Flash-Next-Uncensored-BF16-MRoPE-Fix

leonice2 Qwen MoE multimodal
Your rig guess connected
? Why do I need an app?
Reading your rig…

This is a rough estimate. Install the free app - we'll show exact numbers.

Reading real hardware from your app right now. Numbers below are exact.

Below is the per-quantization compatibility for this model.

curl -H "Authorization: Bearer $ABL_KEY" \
     "https://abliteration.org/api/v1/models/leonice2%2FQwen3.8-Flash-Next-Uncensored-BF16-MRoPE-Fix"
Response includes
  • classification m1
  • files 146
  • author_summary 3 models
  • readme_text full
10 credits · hourly refresh · ~4 KB payload Get an API key →
Abliteration classifier · v1.0.0
M1
Primary method

Direct removal

No other method signals detected in this model.
Confidence
MEDIUM
Why this label 3 signals
Method inferred from partial signals - repository name, related files, or tag patterns. Producer identity not confirmed; label may sharpen or shift as we gather more evidence.
  • 'abliterated' in name/tags
  • is_gguf=0 (base model)
  • no specific method indicators - defaulting to M1 (most common)
Refusal direction extracted via
Extraction technique

Difference-of-means

Confidence
MEDIUM
Why we say so
primary_method=M1; difference-of-means is the reference extraction for M1/M3 (Arditi 2024)
Downloads · 30-day
149
Likes
0
Model age
1w ago
created 2026-11-11
Downloads over time
Now 149 → from 0 ↑ 0%
055109164 0 on Nov 11 149 on Nov 22 Nov
Nov 11 → Nov 22 · 12 snapshots · spans 11 days

Genealogy 0 direct forks

Full fork graph →

This model's place in the market. Above: what it was derived from. Below: the tree of everything derived from it.

Metadata

License
apache-2.0
Languages
en zh
Tags
transformers safetensors qwen4_exp image-text-to-text abliterated qwen qwen4 qwen3.8 flash-next uncensored ai-red-team red-teaming

Related

Control vectors

Total size
335 GB
Files
146
Quantizations
1
Registered
2026-11-11 21:03
Last updated on HF
2026-11-18 02:13

Files by quantization

Auxiliary files 146 files 335 GB
model-00062-of-00131.safetensors 3.27 GB 44382d57 download
model-00060-of-00131.safetensors 3.23 GB 32bd7337 download
model-00005-of-00131.safetensors 3.13 GB c27dfdda download
model-00038-of-00131.safetensors 3.13 GB 063965d1 download
model-00040-of-00131.safetensors 3.13 GB ed06bdec download
model-00042-of-00131.safetensors 3.13 GB 071bb12f download
model-00044-of-00131.safetensors 3.13 GB bab6f9fd download
model-00046-of-00131.safetensors 3.13 GB f838e842 download
model-00048-of-00131.safetensors 3.13 GB dbbf3a6e download
model-00050-of-00131.safetensors 3.13 GB 6014a2b7 download
model-00052-of-00131.safetensors 3.13 GB 1914762c download
model-00054-of-00131.safetensors 3.13 GB 22daa762 download
model-00056-of-00131.safetensors 3.13 GB 86ef0b53 download
model-00061-of-00131.safetensors 3.13 GB cf5bff0a download
model-00063-of-00131.safetensors 3.13 GB e35a86ae download
model-00065-of-00131.safetensors 3.13 GB 040b7305 download
model-00067-of-00131.safetensors 3.13 GB 9223da6f download
model-00069-of-00131.safetensors 3.13 GB f7d460aa download
model-00071-of-00131.safetensors 3.13 GB 184e042c download
model-00073-of-00131.safetensors 3.13 GB 19de39ab download
model-00075-of-00131.safetensors 3.13 GB fc43b07e download
model-00077-of-00131.safetensors 3.13 GB a75bf91d download
model-00079-of-00131.safetensors 3.13 GB e4cd07a1 download
model-00083-of-00131.safetensors 3.13 GB f0971292 download
model-00085-of-00131.safetensors 3.13 GB 69083ff9 download
model-00087-of-00131.safetensors 3.13 GB 9b9b8704 download
model-00089-of-00131.safetensors 3.13 GB 640ea881 download
model-00091-of-00131.safetensors 3.13 GB 6ff4b971 download
model-00093-of-00131.safetensors 3.13 GB 3d4f44c4 download
model-00095-of-00131.safetensors 3.13 GB 9e680b69 download
model-00097-of-00131.safetensors 3.13 GB 19be7659 download
model-00099-of-00131.safetensors 3.13 GB 21daceed download
model-00101-of-00131.safetensors 3.13 GB 2a596073 download
model-00105-of-00131.safetensors 3.13 GB 4040abee download
model-00107-of-00131.safetensors 3.13 GB 08fb87b5 download
model-00109-of-00131.safetensors 3.13 GB 67252c3e download
model-00111-of-00131.safetensors 3.13 GB fe316b24 download
model-00113-of-00131.safetensors 3.13 GB ef412453 download
model-00115-of-00131.safetensors 3.13 GB b536b533 download
model-00117-of-00131.safetensors 3.13 GB 27b091d8 download
model-00119-of-00131.safetensors 3.13 GB fae55500 download
model-00002-of-00131.safetensors 3.13 GB 1da92de7 download
model-00004-of-00131.safetensors 3.13 GB 18b75622 download
model-00058-of-00131.safetensors 3.13 GB b7d2ab70 download
model-00081-of-00131.safetensors 3.13 GB 4cf05b67 download
model-00103-of-00131.safetensors 3.13 GB cc197d5c download
model-00121-of-00131.safetensors 3.13 GB e00e7998 download
model-00123-of-00131.safetensors 3.13 GB bb8e5b70 download
model-00125-of-00131.safetensors 3.13 GB b91a6439 download
model-00127-of-00131.safetensors 3.13 GB fa0c6099 download
model-00129-of-00131.safetensors 3.13 GB d24efe73 download
model-00031-of-00131.safetensors 2.98 GB 9b6bf01d download
model-00032-of-00131.safetensors 2.98 GB 7453fcb5 download
model-00033-of-00131.safetensors 2.98 GB 5b2f577c download
model-00034-of-00131.safetensors 2.98 GB 35b55226 download
model-00035-of-00131.safetensors 2.98 GB 4fe61121 download
model-00036-of-00131.safetensors 2.98 GB 237789c0 download
model-00006-of-00131.safetensors 2.98 GB 8277a48e download
model-00007-of-00131.safetensors 2.98 GB 0fba3693 download
model-00008-of-00131.safetensors 2.98 GB 024f59b6 download
model-00009-of-00131.safetensors 2.98 GB 23fbef15 download
model-00010-of-00131.safetensors 2.98 GB b3a30352 download
model-00011-of-00131.safetensors 2.98 GB d1215943 download
model-00012-of-00131.safetensors 2.98 GB f63d649e download
model-00013-of-00131.safetensors 2.98 GB 5e0f0fb3 download
model-00014-of-00131.safetensors 2.98 GB 5ee73bd2 download
model-00015-of-00131.safetensors 2.98 GB b35421bf download
model-00016-of-00131.safetensors 2.98 GB 73c4b1bd download
model-00017-of-00131.safetensors 2.98 GB ee60c126 download
model-00018-of-00131.safetensors 2.98 GB 72cf3696 download
model-00019-of-00131.safetensors 2.98 GB bdad649b download
model-00020-of-00131.safetensors 2.98 GB 2258dad5 download
model-00021-of-00131.safetensors 2.98 GB 775952ac download
model-00022-of-00131.safetensors 2.98 GB 7d205070 download
model-00023-of-00131.safetensors 2.98 GB 598d4f77 download
model-00024-of-00131.safetensors 2.98 GB 09e53494 download
model-00025-of-00131.safetensors 2.98 GB 482d18f0 download
model-00026-of-00131.safetensors 2.98 GB ac479fe2 download
model-00027-of-00131.safetensors 2.98 GB 4981d4e6 download
model-00028-of-00131.safetensors 2.98 GB 1b45a957 download
model-00029-of-00131.safetensors 2.98 GB 40d2f229 download
model-00030-of-00131.safetensors 2.98 GB 893a9b59 download
model-00130-of-00131.safetensors 2.77 GB 4340c28a download
model-00120-of-00131.safetensors 1.81 GB 9986a2ba download
model-00086-of-00131.safetensors 1.81 GB 4ca23bbf download
model-00041-of-00131.safetensors 1.81 GB a5408f34 download
model-00057-of-00131.safetensors 1.81 GB 205057d9 download
model-00076-of-00131.safetensors 1.81 GB 4edbda09 download
model-00049-of-00131.safetensors 1.80 GB 7cb2969c download
model-00082-of-00131.safetensors 1.80 GB 9a87f0eb download
model-00112-of-00131.safetensors 1.80 GB 505a81b6 download
model-00102-of-00131.safetensors 1.80 GB 0f5d7586 download
model-00094-of-00131.safetensors 1.80 GB 2aad85b7 download
model-00068-of-00131.safetensors 1.80 GB 821efdcf download
model-00126-of-00131.safetensors 1.80 GB bd0defea download
model-00108-of-00131.safetensors 1.77 GB 9711879e download
model-00106-of-00131.safetensors 1.74 GB 39529170 download
model-00072-of-00131.safetensors 1.71 GB fa5ac192 download
model-00053-of-00131.safetensors 1.71 GB e2fbd70e download
model-00088-of-00131.safetensors 1.71 GB 153520d7 download
model-00043-of-00131.safetensors 1.71 GB 2fb07fd5 download
model-00090-of-00131.safetensors 1.71 GB c168d860 download
model-00078-of-00131.safetensors 1.71 GB 2d7b0f81 download
model-00070-of-00131.safetensors 1.71 GB 89c50282 download
model-00114-of-00131.safetensors 1.71 GB 5bb0a685 download
model-00045-of-00131.safetensors 1.71 GB 173e1ae7 download
model-00051-of-00131.safetensors 1.71 GB d68189db download
model-00064-of-00131.safetensors 1.71 GB 72b19eea download
model-00096-of-00131.safetensors 1.71 GB d7c6b292 download
model-00098-of-00131.safetensors 1.71 GB 2a218469 download
model-00116-of-00131.safetensors 1.71 GB 869b7359 download
model-00104-of-00131.safetensors 1.71 GB 06b883b0 download
model-00003-of-00131.safetensors 1.71 GB 2c99790f download
model-00122-of-00131.safetensors 1.71 GB 73ab290a download
model-00128-of-00131.safetensors 1.71 GB 094e047e download
model-00037-of-00131.safetensors 1.62 GB 0102d1d6 download
model-00124-of-00131.safetensors 1.61 GB a3119e1f download
model-00066-of-00131.safetensors 1.60 GB 1243856f download
model-00047-of-00131.safetensors 1.60 GB f0d2d092 download
model-00118-of-00131.safetensors 1.60 GB ecc03246 download
model-00100-of-00131.safetensors 1.60 GB 3d82bfe1 download
model-00059-of-00131.safetensors 1.60 GB 29304fa1 download
model-00039-of-00131.safetensors 1.60 GB 42dbb0fc download
model-00055-of-00131.safetensors 1.60 GB 66fc80b8 download
model-00074-of-00131.safetensors 1.60 GB fca05ba3 download
model-00080-of-00131.safetensors 1.60 GB 14b5bad0 download
model-00084-of-00131.safetensors 1.60 GB c83d4e92 download
model-00092-of-00131.safetensors 1.60 GB e72019cd download
model-00110-of-00131.safetensors 1.60 GB e5968bf4 download
model-00131-of-00131.safetensors 1.18 GB 50be0ccd download
model-00001-of-00131.safetensors 992 MB 204f4b70 download
tokenizer.json 12.2 MB 0997f410 download
vocab.json 6.41 MB 0aa0ce06 download
merges.txt 3.20 MB a494e019 download
model.safetensors.index.json 167 KB a234b02a download
tokenizer_config.json 17.5 KB 5de744b3 download
README.md 13.4 KB 5cddde6e download
chat_template.jinja 8.74 KB c0c686f9 download
config.json 4.63 KB 491017e9 download
mrope_startup_plugin.py 4.34 KB be58fd22 download
LICENSE 3.16 KB 9557a896 download
.gitattributes 1.53 KB 52373fe2 download
DEPLOYMENT.txt 454 B b643de11 download
preprocessor_config.json 390 B 2ea84a43 download
video_preprocessor_config.json 385 B 3ba673a5 download
generation_config.json 202 B 023756cf download

README current version from Hugging Face


license: apache-2.0
base_model: Qwen/Qwen3.8-Flash-Next
base_model_relation: finetune
pipeline_tag: image-text-to-text
library_name: transformers
language:

  • en
  • zh
    tags:
  • abliterated
  • qwen
  • qwen4
  • qwen3.8
  • flash-next
  • uncensored
  • ai-red-team
  • red-teaming
  • moe
  • vision-language
  • function-calling
  • reasoning
  • mtp
  • bf16
  • post-training
  • fine-tuning

OrcaRouter

Qwen3.8-Flash-Next-Uncensored

The full-precision BF16 abliterated (refusal-removed) build of Qwen's Qwen3.8-Flash-Next — the source for fine-tuning & quantization

OrcaCyber Zero 1.5

🛡️ New from Orca — OrcaCyber Zero 1.5

Post-trained for cybersecurity: vulnerability analysis, vulnerability reproduction, red teaming & offensive-security research. 98.01% pass@1 on CyberGym Level 1 · 1M context · function calling · OpenAI-compatible API.

Try it on OrcaRouter

🔌 New from Orca — Orca Connect

Serving this with vLLM? Orca Connect puts the rig under your desk behind your usual OrcaRouter API key, next to 200+ hosted models, reachable from a laptop, a phone or a cloud agent. Outbound only: no port forwarding, no tunnel, no firewall rule. Requests your own hardware serves cost $0, and orca-connect claude points Claude Code at your rig for one session.

Connect your rig

Website Model Catalog License precision 262K context Vision-Language MoE MTP

One Gateway. Every Model. — Route Smarter · Ship Safer · Spend Less.

Website · Model Catalog · GitHub · Discord · X


The full-precision BF16 abliterated (refusal-removed) build of Qwen/Qwen3.8-Flash-Next — a ~180B Mixture-of-Experts (125B params / ~6B active), hybrid-attention, native vision-language model with flexible thinking, tool-calling, and an MTP head. These are the source weights from which the quantized releases are derived, and the recommended base for further fine-tuning / post-training and quantization — the full vision tower and MTP head are preserved.

Derived releases:  •  Qwen3.8-Flash-Next-Uncensored (BF16 source)  •  Qwen3.8-Flash-Next-Uncensored-FP8 (block-FP8, mirrors official)  •  Qwen3.8-Flash-Next-Uncensored-MLX (4 / 6 / 8-bit, Apple Silicon).


⚠️ Disclaimer — read before use

This model has had its safety alignment substantially removed via abliteration (orthogonalizing the
refusal direction out of the residual stream). As a direct consequence:

  • It will comply with harmful, unethical, offensive, or illegal requests that the original
    Qwen3.8-Flash-Next would refuse. It has no meaningful built-in guardrails.
  • It is released strictly for legitimate research — interpretability, AI-safety and refusal-mechanism
    study, red-teaming, robustness evaluation, and controlled experiments.
  • You assume full responsibility and liability for how you use it and for everything it generates. Do
    not deploy it to end users or in production without adding your own safety, moderation, and
    abuse-prevention layers.
  • Use must comply with the Apache 2.0 License inherited
    from the base model, and all laws and regulations that apply to you.
  • The authors and uploaders accept no liability for any misuse or harm. Its outputs do not reflect
    the views of the uploaders or of Qwen / Alibaba.

By downloading or using this model you acknowledge and accept the above.


🐋 Run it with OrcaCode Review

Models are only half the system.

OrcaCode Review turns every model listed on
OrcaRouter into a production code-review agent:

  • reviews every PR
  • finds security + correctness issues
  • posts inline findings
  • P0/P1 can block merges
  • swap models anytime

Open model. Open harness. Open bill.


Model details

Base model Qwen/Qwen3.8-Flash-Next
Architecture Qwen4ExpForConditionalGeneration (qwen4_exp, Qwen4 preview) — 48 layers, hidden 2560, hybrid attention (36 Gated-DeltaNet linear + 12 full-attention, interval 4), 512 fused experts, top-10 + shared expert, 51B-param PLE n-gram embedding, Hyper-Connections residual, native vision + video tower, and an MTP speculative-decoding head
Modification Abliteration (refusal-direction removal) on the BF16 weights — no quantization
Format safetensors, BF16, 131 shards (336 GB, 1658 tensors)
Preserved Full vision + video tower (333 visual.*) and MTP head (31 mtp.*)
Context 262,144 tokens
Recommended for Fine-tuning / post-training (SFT · DPO · RL), re-quantization, interpretability & red-team research

Abliteration

Refusal-direction removal following Arditi et al. (2024), Refusal in Language Models Is Mediated by a
Single Direction
. A single refusal direction r (k = 1) is estimated from the block-input residual
stream
(the 2560-d Hyper-Connections GR-Read output, where the refusal direction is linearly separable —
the widened 4-branch output_hidden_states smears it) as the massive-activation-masked mean-difference of
harmful − harmless activations, selected at layer 24 by a full 9-layer quality sweep
(harmful 0.00 / KL 0.085). r is then orthogonalized out of every residual-writing matrix —
W' = W − r(rᵀW) — computed in float32:

Component matrices edited
self_attn.o_proj (12 full-attention layers + MTP) 13
linear_attn.out_proj (36 GDN linear-attention layers) 36
mlp.experts.down_proj (fused 3D, all 512 experts × 49 layers) 49
mlp.shared_expert.down_proj 49
ple.value_proj + embed_tokens (row space) 2
Total residual-writer tensors 149

Preserved (never touched): the full vision + video tower (333 visual.* tensors), the MoE
router (mlp.gate), the fused experts.gate_up_proj reader, all Hyper-Connection mixers, the
QSA sparse-attention indexer, the n-gram embedding table, mtp.fc_*, norms, and lm_head. The MTP
head's residual writers are abliterated consistently so speculative decoding keeps working. Max residual
leakage after the edit: 0.0755 (float32 projection → bf16 storage).

This is a surgical weight edit — it changes ~0 general capability (see Evaluation) while collapsing
refusal behaviour.


Fine-tuning & post-training

This BF16 checkpoint is the recommended base for post-training — full precision, whole VL tower + MTP
head, a drop-in for Qwen/Qwen3.8-Flash-Next in the transformers / Qwen4ExpForConditionalGeneration
stack (needs transformers 5.16+).

  • SFT / DPO / RL work as on the base model (full-FT or LoRA/QLoRA).
  • Re-quantization: quantize from these weights to FP8 / MLX / etc. (this is exactly how the derived
    releases were produced).
  • Note: abliteration is a weight edit, not data-level unlearning. Fine-tuning on refusal-heavy /
    safety data can partially re-introduce refusals; neutral / task data preserves the uncensored behaviour.

Intended use

  • Research into refusal mechanisms, alignment, and interpretability.
  • Red-teaming and safety / robustness evaluation in controlled environments.
  • A full-precision base for further fine-tuning, post-training, and quantization.

Evaluation

Measured on these exact weights served with vLLM (day-0 qwen4_exp image), against the official Qwen/Qwen3.8-Flash-Next served identically with the same scripts and settings. Full BF16. Refusal is judged by a rule-based opening-phrase classifier (caveat = answered but wrapped in a disclaimer) — indicative, not an LLM-judge / publication-grade number. Thinking is off unless stated.

Safety — harmful-prompt refusal, thinking OFF (lower = more uncensored)

Benchmark n Base This model Caveat (this)
AdvBench 100 100.0% 2.0% 53.0%
JailbreakBench (harmful) 100 94.0% 0.0% 52.0%
StrongREJECT 150 99.3% 1.3% 62.0%
HarmBench (standard) 150 100.0% 1.3% 50.7%
MaliciousInstruct 100 98.0% 0.0% 55.0%
SimpleSafetyTests 50 64.0% 2.0% 68.0%
ForbiddenQuestions 150 75.3% 3.3% 57.3%
Custom probes (zh/en) 11 63.6% 0.0% 36.4%

Safety — harmful-prompt refusal, thinking ON (enable_thinking=true)

Judged on the final answer (text after </think>; responses truncated before a final answer are excluded):

thinking-ON refusal Base This model
harmful prompts ~100% ~0%

Over-refusal — benign prompts wrongly refused (lower = better)

Benchmark n Base (no-think / think) This model (no-think / think)
XSTest-safe 250 9.6% / 0.4% 1.2% / 0.4%

Capability retention — vs the official base (same scripts, same settings)

Benchmark n Base This model Δ
MMLU (0-shot letter) 300 90.0% 87.7% -2.3
MMLU-Pro (CoT) 400 77.8% 76.2% -1.5
GSM8K (CoT) 150 92.0% 93.3% +1.3
CMMLU (0-shot, Chinese) 500 81.8% 81.6% -0.2

Verified working on this build: tool-calling ✅, vision/OCR ✅ 6/6, reasoning ✅. The vision + video tower is preserved byte-for-byte (333 visual.* tensors in BF16), so it remains a full vision-language model — a drop-in for the base.


Usage — self-host with vLLM (OpenAI-compatible)

qwen4_exp needs the day-0 vLLM image and transformers 5.16+.

docker run -d --name flashnext --gpus all --ipc host -p 8000:8000 \
  -v /path/to/Qwen3.8-Flash-Next-Uncensored:/model \
  vllm/vllm-openai:qwen38-flash-next-x86_64-cu130 \
  --model /model --served-model-name Qwen3.8-Flash-Next-Uncensored \
  --tensor-parallel-size 8 --trust-remote-code --max-model-len 262144 \
  --enable-expert-parallel --enable-auto-tool-choice --tool-call-parser qwen3_coder

--enable-expert-parallel is required for the FP8 build (MoE intermediate 640 ÷ TP is not divisible by the FP8 block 128 without it); it is harmless for BF16. Tool calls use the Qwen3-Coder XML format (--tool-call-parser qwen3_coder). Toggle reasoning per
request with chat_template_kwargs={"enable_thinking": true|false}; pass image_url content parts for
vision.

Bias, risks, and limitations

  • Safety guardrails removed — the model will produce harmful, biased, or offensive content on request
    (see the disclaimer).
  • It inherits any biases and limitations of the base Qwen3.8-Flash-Next.
  • The reported refusal metric is a rule-based heuristic; evaluate rigorously for your own use case.

License

Apache 2.0, inherited from the base model
Qwen/Qwen3.8-Flash-Next. Abliteration does not change
the underlying license obligations.

README history 1 version

The author's README evolved over time. Click a version to see its content at that point.

  1. 2026-11-11 Copy files from models/orcarouter/Qwen3.8-Flash-Next-Uncensored 4c46e79 13.4 KB
    Loading...
Catalog is the map. Apps are the tools.

Run models on your own machine, not in the cloud.

Every model page has an "Open in Abliteration" button that hands the model directly to the first-party desktop client, at the quantization your rig can actually run. No API keys, no subscription, no prompt leakage.

Open in Abliteration